Math. Appl. 4 (2015), 1{16 DOI: 10.13164/ma.2015.05 CONTINUOUS FUNCTIONS OVER DISCRETE PARTIAL ORDERS JOHN L. PFALTZ Abstract. This paper examines the properties of structure preserving morphisms f over discrete partial orders. It employs concepts of continuity and path homomor- phisms. It will conclude that no single constraint on f will be sufficient, and it will also conclude that a convexity constraint on f −1 seems to be essential. We employ closure lattices to help reach this conclusion. 1. Introduction Mathematics is hugely concerned with functions that preserve the essential struc- ture of its mathematical objects even as they change specific details. We think of continuous functions, f : Rm ! Rn; of group and ring homomorphisms; and of automorphism concepts. But, what constitutes a \structure preserving" morphism of a partially ordered set has not received as much study as these more familiar examples. This is sur- prising, because partial orders, or partially ordered sets, or posets, which are terms that we will use interchangeably, are common in applied mathematics. They are ubiquitous in computer science. They are readily a model of concurrent processing [3,7], and any system of events occurring in time, such as task scheduling [8]. The theory of data types [9], and abstract models of computation, such as denotational semantics [33] have been extensively studied as \directed-complete partial orders", or dcpo's, in domain theory [1, 16, 37]. In \formal concept analysis", or FCA [15], the concepts have a partially ordered structure. There is no shortage of application instances. In pure mathematics, lattice theory [4, 5, 18] is the study of \well-behaved" partial orders; they are fundamental in some algebraic systems, such as \Stone Spaces" [21]. So, partial orders are widely used and very well understood { except for their functional change. Functions that map one partial order into another are far less easy to characterize. 1.1. Basic terminology Partially ordered sets, denoted (P; ≤), are a familiar mathematical system. The element ordering, ≤, is a pre-order if for all elements x; y; z 2 P we have MSC (2010): primary 06A06, 06A11. Keywords: partial order, transformation, homomorphism, convex subgraph inverse map. 1 2 J. L. PFALTZ (A1) x ≤ x reflexive (A2) x ≤ y and y ≤ z imply that x ≤ z, transitive If in addition (A3) x ≤ y and y ≤ x imply that x = y, anti-symmetric1 then (P; ≤) is a partially ordered set. Usually these three axiomatic properties are presented together; but we shall see that (A3) is very different in character from (A2). While it is convenient to have ≤ be reflexive (A1); nothing essential is changed if it is omitted, or if the order is assumed to be \anti-reflexive" with x 6≤ x. Let (P; ≤) be a partially ordered set. An element y is said to be a maximal element of S ⊆ P if y 2 S and there exists no element z 2 S; z 6= y such that y ≤ z. An element y is said to be an upper bound, or u:b:, of S ⊆ P if for all x 2 S, x ≤ y. An element y is said to be a least upper bound, l:u:b, of S if it is an upper bound and if for all upper bounds zi of S, y ≤ zi. Minimal elements, lower bounds (l:b:) and greatest lower bounds (g:l:b:) are defined analogously. A subset S ⊆ P is said to be totally ordered if for all x; y 2 S, x 6= y implies either x < y or y < x.A chain, C(x; z) ⊆ P is a totally ordered subset C with minimal element x 2 C and maximal element z 2 C. A partial order (P; ≤) is said to be discrete if for every chain C(x; z), the set C(x; z)nfzg is a chain with a maximal element. (For this paper, we also assume that C(x; z)nfxg is a chain with a minimal element.) This distinguishes our approach to partial orders from that of domain theory [1, 16] which is largely concerned with convergence and approximation of least upper bounds in infinite partial orders. Any finite partial order must be discrete. Discrete partial orders need not be finite. Because we restrict ourselves to discrete partial orders, we can illustrate the concepts of this paper with directed graphs. A chain, C(x; z) in (P; ≤) can be visualized as a path ρ(x; z) from x to z, consisting of a sequence of elements < x; : : : ; yi; : : : ; z > where x ≤ yi ≤ z. We shall use :ρ to denote the non-existence of any path. Thus x :ρ y is equivalent to x 6≤ y. By the length of ρ(x; z), which we denote by jρ(x; z)j, we mean the number of elements in the chain C(x; z) less one. Thus, every element y has a path ρ(y; y) of length zero. If y = max(C(x; z)nfzg), we say z covers y. Equivalently, we could say there is a \directed edge" between y and z.2 Figure 1 illustrates a partial order ≤ on P = fa; : : : hg, in which g covers e. We choose to represent partial orders in a horizontal (left to right) manner rather than the more conventional vertical Hasse diagram because it conserves space on the printed page. The arrow heads suggest the ≤ relation with \lesser" elements to the left. As with the ≤ relation, we will use x ρ z to denote that there exists at least one path from x to z, and let ρ(x; z) denote a specific path, in much the same way that a chain denotes a specific total ordering. Those readers more comfortable with the order terminology using ≤ may replace every occurrence of ρ with ≤. 1There are various names for (A3); some would call this \weak anti-symmetry" and want \anti-symmetry" to mean that x < y implies y 6< x. 2It is customary to define graphs as an \edge" relation E on the ground set of elements P , e.g., [2, 11, 19, 26]. Since our focus is on transitive ordering relations it seems more natural to emphasize the \path" relation, and let edges denote covering situations. CONTINUOUS FUNCTIONS OVER DISCRETE PARTIAL ORDERS 3 c e a g d f h b Figure 1. A representative partially ordered set (P; ρ). 3 A cycle is a path ρ(x; : : : yi; : : : x) of length ≥ 2. A directed graph (P; ρ) is said to be acyclic if it contains no cycles. It is well known that Proposition 1.1. A discrete pre-ordered set (P; ≤) is a partial order, if and only if its corresponding directed graph (P; ρ) is acyclic. 2. Order preserving functions A function f :(P; ≤) ! (P 0; ≤0) is said to be order preserving , or isotone, if x ≤ z implies f(x) ≤0 f(z). In our discrete case this is equivalent to saying x ρ z implies f(x) ρ0 f(z).4 However, order/path preserving functions need not preserve partial orders. This is most easily demonstrated when we visualize the partial orders as acyclic graphs. Let (P; ρ) be that of Figure 1. Let f(b) = f(h) = b0 = h0 and let f(x) = x0 for x = fa; c; d; e; f; gg as in Figure 2. While f is readily path preserving, c e c’ e’ a g f a’ g’ d f h d’ f’ b’= h’ b Figure 2. An order/path preserving function on Figure 1. (P 0; ρ0) is not acyclic; it is not a partial order. 2.1. Categorical posets In some texts [10, 31], the category, Poset, of partially ordered sets is defined to consist of the class of all partially ordered sets as its objects, and the class of all f isotone functions [A −! B] between objects A and B. It is easy to understand why this is often presented as an early example; it is not hard to show that the essential axioms of a category are satisfied, namely that the composition f·g, of isotone functions f and g, is isotone; that identity functions idA and idB exist for all objects A; B; and that idA·f and f·idB are isotone. But, does this category really characterize partial orders? It does characterize a preorder, which is any transitive order relation; but it does not capture the \antisymmetry" aspect. This is tacitly acknowledged in [1] where we have \If we drop antisymmetry from our list of requirement than we get what is known as preorders. This does not change 3In definition of \path" we assumed a path of length 0 at every node to assure reflexivity; consequently \loops" or edges of the form (x; x) are simply ignored, if they exist. 4It is correct to denote the order/path relation on P 0 by ≤0 and ρ0, because the ordering relation may not be the same as on P . 4 J. L. PFALTZ the theory very much." Mac Lane [24] never defines a poset category. The author knows of no definition of a P oset category that is distinct from that of a P reOrder or T otalOrder category; but see [36], which is quite relevant to this paper. 2.2. Graph theoretic maps In graph theory, the term \homomorphism" is widely used to describe an \edge preserving" function f : G ! G0 where G = (P; E), G0 = (P 0;E0) and (x; y) 2 E implies (f(x); f(y)) 2 E0, [17, 26]. A function f satisfying this definition of \homomorphism" is path preserving, or order preserving.